LearnNewsExamplesServices
Frontmatter
id17042
titleFriction→gold: provider-lanes epic retrospective — convert the team-failure mechanics into substrate
stateOpen
labels
aiagent-os
assigneesneo-opus-vega
createdAtAug 12, 2026, 11:47 PM
updatedAtAug 14, 2026, 10:55 AM
githubUrlhttps://github.com/neomjs/neo/issues/17042
authorneo-opus-vega
commentsCount1
parentIssuenull
subIssues[]
subIssuesCompleted0
subIssuesTotal0
contentTrust
projected
quarantined0
signals[]
blockedBy[]
blocking[]

Friction→gold: provider-lanes epic retrospective — convert the team-failure mechanics into substrate

neo-opus-vega
neo-opus-vega commented on Aug 12, 2026, 11:47 PM

Context

2026-08-12: epic #17018 (role-isolated provider lanes) was a finish-today goal serving an external deployment deadline whose existential stakes were clarified before the source Discussion. The day produced roughly 19k+ inserted lines across six-plus PRs and burned multiple weekly review flatrates — while the deployment's two failing symptoms (four permanently pinned CPU cores; ingestion halting and never resuming) remained the only outcomes that counted. The operator names the day a possible gigantic team failure and directs the learnings into friction→gold. The larger operator-named pattern: the external plane has never worked since May across 3000+ merged PRs — internal artifact velocity without client-outcome delivery.

The mechanics to convert into substrate (each → one bounded change)

  1. Decision-owner criterion check. The {1,2,4} "smallest slot count passing a joint SLO" election criterion was invented inside the Discussion fold (the author folding a peer falsifier); the decision owner had already decided slots=4 from production parity, and the only open requirement was freeze-safety. Substrate: before any instrument, benchmark, or election is designed, the design must name the decision owner and quote their criterion — a criterion without an owner is a research question, not an AC. Candidate home: ideation-sandbox workflow (fold gate) + epic-create.
  2. Client-outcome gate for deadline epics. Every leaf of an externally-deadlined epic must trace to the failing symptom it shortens; leaves that do not shorten red→green sequence behind the release by default. Candidate home: epic-create workflow.
  3. Simplicity ownership in review. Review culture answered every falsifier by adding mechanism; no reviewer role owns "is the machine bigger than its decision?". Candidate home: pr-review guide — a required proportionality check with refusal power, coordinated with #17036 (recurrence tripwire) and #17037 (right-sizing audit) so the three do not overlap.
  4. Instrument fossilization. The election runner → runtime proof → cutover manifest each hard-require the previous instrument's full ceremony artifact (validators hardcode three candidate receipts and ELECTED status), so retiring one decision bricked all three instruments the same night — including for the purpose they were nominally built to serve. Substrate: instruments consume a decision record (values + provenance), never each other's full ceremony artifacts.

Acceptance Criteria

  • Each mechanic lands as one bounded substrate PR, or a recorded rejection with rationale — net-loaded-bytes discipline applies to every change.
  • The retrospective summary posts on epic #17018's closeout before that epic closes.
  • Sequencing: nothing here runs before the external release ships; this ticket must not add process weight to the release path.

Out of Scope

  • Re-litigating the epic's merged code (#17036/#17037 own the code-side debt).
  • Any client-identifying detail (public artifact).

Related

Epic #17018 · D#17015 · #17036 · #17037

Origin Session ID: 379c88ee-52c5-41ad-8973-8f28ebc8cbd6